1 /* SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) */
2 /* Copyright (C) 2015-2018 Netronome Systems, Inc. */
3
4 /*
5 * nfp_net.h
6 * Declarations for Netronome network device driver.
7 * Authors: Jakub Kicinski <jakub.kicinski@netronome.com>
8 * Jason McMullan <jason.mcmullan@netronome.com>
9 * Rolf Neugebauer <rolf.neugebauer@netronome.com>
10 */
11
12 #ifndef _NFP_NET_H_
13 #define _NFP_NET_H_
14
15 #include <linux/atomic.h>
16 #include <linux/interrupt.h>
17 #include <linux/list.h>
18 #include <linux/netdevice.h>
19 #include <linux/pci.h>
20 #include <linux/dim.h>
21 #include <linux/io-64-nonatomic-hi-lo.h>
22 #include <linux/semaphore.h>
23 #include <linux/workqueue.h>
24 #include <net/xdp.h>
25
26 #include "nfp_net_ctrl.h"
27
28 #define nn_pr(nn, lvl, fmt, args...) \
29 ({ \
30 struct nfp_net *__nn = (nn); \
31 \
32 if (__nn->dp.netdev) \
33 netdev_printk(lvl, __nn->dp.netdev, fmt, ## args); \
34 else \
35 dev_printk(lvl, __nn->dp.dev, "ctrl: " fmt, ## args); \
36 })
37
38 #define nn_err(nn, fmt, args...) nn_pr(nn, KERN_ERR, fmt, ## args)
39 #define nn_warn(nn, fmt, args...) nn_pr(nn, KERN_WARNING, fmt, ## args)
40 #define nn_info(nn, fmt, args...) nn_pr(nn, KERN_INFO, fmt, ## args)
41 #define nn_dbg(nn, fmt, args...) nn_pr(nn, KERN_DEBUG, fmt, ## args)
42
43 #define nn_dp_warn(dp, fmt, args...) \
44 ({ \
45 struct nfp_net_dp *__dp = (dp); \
46 \
47 if (unlikely(net_ratelimit())) { \
48 if (__dp->netdev) \
49 netdev_warn(__dp->netdev, fmt, ## args); \
50 else \
51 dev_warn(__dp->dev, fmt, ## args); \
52 } \
53 })
54
55 /* Max time to wait for NFP to respond on updates (in seconds) */
56 #define NFP_NET_POLL_TIMEOUT 5
57
58 /* Interval for reading offloaded filter stats */
59 #define NFP_NET_STAT_POLL_IVL msecs_to_jiffies(100)
60
61 /* Bar allocation */
62 #define NFP_NET_CTRL_BAR 0
63 #define NFP_NET_Q0_BAR 2
64 #define NFP_NET_Q1_BAR 4 /* OBSOLETE */
65
66 /* Max bits in DMA address */
67 #define NFP_NET_MAX_DMA_BITS 40
68
69 /* Default size for MTU and freelist buffer sizes */
70 #define NFP_NET_DEFAULT_MTU 1500U
71
72 /* Maximum number of bytes prepended to a packet */
73 #define NFP_NET_MAX_PREPEND 64
74
75 /* Interrupt definitions */
76 #define NFP_NET_NON_Q_VECTORS 2
77 #define NFP_NET_IRQ_LSC_IDX 0
78 #define NFP_NET_IRQ_EXN_IDX 1
79 #define NFP_NET_MIN_VNIC_IRQS (NFP_NET_NON_Q_VECTORS + 1)
80
81 /* Queue/Ring definitions */
82 #define NFP_NET_MAX_TX_RINGS 64 /* Max. # of Tx rings per device */
83 #define NFP_NET_MAX_RX_RINGS 64 /* Max. # of Rx rings per device */
84 #define NFP_NET_MAX_R_VECS (NFP_NET_MAX_TX_RINGS > NFP_NET_MAX_RX_RINGS ? \
85 NFP_NET_MAX_TX_RINGS : NFP_NET_MAX_RX_RINGS)
86 #define NFP_NET_MAX_IRQS (NFP_NET_NON_Q_VECTORS + NFP_NET_MAX_R_VECS)
87
88 #define NFP_NET_MIN_TX_DESCS 256 /* Min. # of Tx descs per ring */
89 #define NFP_NET_MIN_RX_DESCS 256 /* Min. # of Rx descs per ring */
90 #define NFP_NET_MAX_TX_DESCS (256 * 1024) /* Max. # of Tx descs per ring */
91 #define NFP_NET_MAX_RX_DESCS (256 * 1024) /* Max. # of Rx descs per ring */
92
93 #define NFP_NET_TX_DESCS_DEFAULT 4096 /* Default # of Tx descs per ring */
94 #define NFP_NET_RX_DESCS_DEFAULT 4096 /* Default # of Rx descs per ring */
95
96 #define NFP_NET_FL_BATCH 16 /* Add freelist in this Batch size */
97 #define NFP_NET_XDP_MAX_COMPLETE 2048 /* XDP bufs to reclaim in NAPI poll */
98
99 /* Offload definitions */
100 #define NFP_NET_N_VXLAN_PORTS (NFP_NET_CFG_VXLAN_SZ / sizeof(__be16))
101
102 #define NFP_NET_RX_BUF_HEADROOM (NET_SKB_PAD + NET_IP_ALIGN)
103 #define NFP_NET_RX_BUF_NON_DATA (NFP_NET_RX_BUF_HEADROOM + \
104 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
105
106 /* Forward declarations */
107 struct nfp_cpp;
108 struct nfp_eth_table_port;
109 struct nfp_net;
110 struct nfp_net_r_vector;
111 struct nfp_port;
112
113 /* Convenience macro for wrapping descriptor index on ring size */
114 #define D_IDX(ring, idx) ((idx) & ((ring)->cnt - 1))
115
116 /* Convenience macro for writing dma address into RX/TX descriptors */
117 #define nfp_desc_set_dma_addr(desc, dma_addr) \
118 do { \
119 __typeof(desc) __d = (desc); \
120 dma_addr_t __addr = (dma_addr); \
121 \
122 __d->dma_addr_lo = cpu_to_le32(lower_32_bits(__addr)); \
123 __d->dma_addr_hi = upper_32_bits(__addr) & 0xff; \
124 } while (0)
125
126 /* TX descriptor format */
127
128 #define PCIE_DESC_TX_EOP BIT(7)
129 #define PCIE_DESC_TX_OFFSET_MASK GENMASK(6, 0)
130 #define PCIE_DESC_TX_MSS_MASK GENMASK(13, 0)
131
132 /* Flags in the host TX descriptor */
133 #define PCIE_DESC_TX_CSUM BIT(7)
134 #define PCIE_DESC_TX_IP4_CSUM BIT(6)
135 #define PCIE_DESC_TX_TCP_CSUM BIT(5)
136 #define PCIE_DESC_TX_UDP_CSUM BIT(4)
137 #define PCIE_DESC_TX_VLAN BIT(3)
138 #define PCIE_DESC_TX_LSO BIT(2)
139 #define PCIE_DESC_TX_ENCAP BIT(1)
140 #define PCIE_DESC_TX_O_IP4_CSUM BIT(0)
141
142 struct nfp_net_tx_desc {
143 union {
144 struct {
145 u8 dma_addr_hi; /* High bits of host buf address */
146 __le16 dma_len; /* Length to DMA for this desc */
147 u8 offset_eop; /* Offset in buf where pkt starts +
148 * highest bit is eop flag.
149 */
150 __le32 dma_addr_lo; /* Low 32bit of host buf addr */
151
152 __le16 mss; /* MSS to be used for LSO */
153 u8 lso_hdrlen; /* LSO, TCP payload offset */
154 u8 flags; /* TX Flags, see @PCIE_DESC_TX_* */
155 union {
156 struct {
157 u8 l3_offset; /* L3 header offset */
158 u8 l4_offset; /* L4 header offset */
159 };
160 __le16 vlan; /* VLAN tag to add if indicated */
161 };
162 __le16 data_len; /* Length of frame + meta data */
163 } __packed;
164 __le32 vals[4];
165 __le64 vals8[2];
166 };
167 };
168
169 /**
170 * struct nfp_net_tx_buf - software TX buffer descriptor
171 * @skb: normal ring, sk_buff associated with this buffer
172 * @frag: XDP ring, page frag associated with this buffer
173 * @dma_addr: DMA mapping address of the buffer
174 * @fidx: Fragment index (-1 for the head and [0..nr_frags-1] for frags)
175 * @pkt_cnt: Number of packets to be produced out of the skb associated
176 * with this buffer (valid only on the head's buffer).
177 * Will be 1 for all non-TSO packets.
178 * @real_len: Number of bytes which to be produced out of the skb (valid only
179 * on the head's buffer). Equal to skb->len for non-TSO packets.
180 */
181 struct nfp_net_tx_buf {
182 union {
183 struct sk_buff *skb;
184 void *frag;
185 };
186 dma_addr_t dma_addr;
187 short int fidx;
188 u16 pkt_cnt;
189 u32 real_len;
190 };
191
192 /**
193 * struct nfp_net_tx_ring - TX ring structure
194 * @r_vec: Back pointer to ring vector structure
195 * @idx: Ring index from Linux's perspective
196 * @qcidx: Queue Controller Peripheral (QCP) queue index for the TX queue
197 * @qcp_q: Pointer to base of the QCP TX queue
198 * @cnt: Size of the queue in number of descriptors
199 * @wr_p: TX ring write pointer (free running)
200 * @rd_p: TX ring read pointer (free running)
201 * @qcp_rd_p: Local copy of QCP TX queue read pointer
202 * @wr_ptr_add: Accumulated number of buffers to add to QCP write pointer
203 * (used for .xmit_more delayed kick)
204 * @txbufs: Array of transmitted TX buffers, to free on transmit
205 * @txds: Virtual address of TX ring in host memory
206 * @dma: DMA address of the TX ring
207 * @size: Size, in bytes, of the TX ring (needed to free)
208 * @is_xdp: Is this a XDP TX ring?
209 */
210 struct nfp_net_tx_ring {
211 struct nfp_net_r_vector *r_vec;
212
213 u32 idx;
214 int qcidx;
215 u8 __iomem *qcp_q;
216
217 u32 cnt;
218 u32 wr_p;
219 u32 rd_p;
220 u32 qcp_rd_p;
221
222 u32 wr_ptr_add;
223
224 struct nfp_net_tx_buf *txbufs;
225 struct nfp_net_tx_desc *txds;
226
227 dma_addr_t dma;
228 size_t size;
229 bool is_xdp;
230 } ____cacheline_aligned;
231
232 /* RX and freelist descriptor format */
233
234 #define PCIE_DESC_RX_DD BIT(7)
235 #define PCIE_DESC_RX_META_LEN_MASK GENMASK(6, 0)
236
237 /* Flags in the RX descriptor */
238 #define PCIE_DESC_RX_RSS cpu_to_le16(BIT(15))
239 #define PCIE_DESC_RX_I_IP4_CSUM cpu_to_le16(BIT(14))
240 #define PCIE_DESC_RX_I_IP4_CSUM_OK cpu_to_le16(BIT(13))
241 #define PCIE_DESC_RX_I_TCP_CSUM cpu_to_le16(BIT(12))
242 #define PCIE_DESC_RX_I_TCP_CSUM_OK cpu_to_le16(BIT(11))
243 #define PCIE_DESC_RX_I_UDP_CSUM cpu_to_le16(BIT(10))
244 #define PCIE_DESC_RX_I_UDP_CSUM_OK cpu_to_le16(BIT(9))
245 #define PCIE_DESC_RX_DECRYPTED cpu_to_le16(BIT(8))
246 #define PCIE_DESC_RX_EOP cpu_to_le16(BIT(7))
247 #define PCIE_DESC_RX_IP4_CSUM cpu_to_le16(BIT(6))
248 #define PCIE_DESC_RX_IP4_CSUM_OK cpu_to_le16(BIT(5))
249 #define PCIE_DESC_RX_TCP_CSUM cpu_to_le16(BIT(4))
250 #define PCIE_DESC_RX_TCP_CSUM_OK cpu_to_le16(BIT(3))
251 #define PCIE_DESC_RX_UDP_CSUM cpu_to_le16(BIT(2))
252 #define PCIE_DESC_RX_UDP_CSUM_OK cpu_to_le16(BIT(1))
253 #define PCIE_DESC_RX_VLAN cpu_to_le16(BIT(0))
254
255 #define PCIE_DESC_RX_CSUM_ALL (PCIE_DESC_RX_IP4_CSUM | \
256 PCIE_DESC_RX_TCP_CSUM | \
257 PCIE_DESC_RX_UDP_CSUM | \
258 PCIE_DESC_RX_I_IP4_CSUM | \
259 PCIE_DESC_RX_I_TCP_CSUM | \
260 PCIE_DESC_RX_I_UDP_CSUM)
261 #define PCIE_DESC_RX_CSUM_OK_SHIFT 1
262 #define __PCIE_DESC_RX_CSUM_ALL le16_to_cpu(PCIE_DESC_RX_CSUM_ALL)
263 #define __PCIE_DESC_RX_CSUM_ALL_OK (__PCIE_DESC_RX_CSUM_ALL >> \
264 PCIE_DESC_RX_CSUM_OK_SHIFT)
265
266 struct nfp_net_rx_desc {
267 union {
268 struct {
269 u8 dma_addr_hi; /* High bits of the buf address */
270 __le16 reserved; /* Must be zero */
271 u8 meta_len_dd; /* Must be zero */
272
273 __le32 dma_addr_lo; /* Low bits of the buffer address */
274 } __packed fld;
275
276 struct {
277 __le16 data_len; /* Length of the frame + meta data */
278 u8 reserved;
279 u8 meta_len_dd; /* Length of meta data prepended +
280 * descriptor done flag.
281 */
282
283 __le16 flags; /* RX flags. See @PCIE_DESC_RX_* */
284 __le16 vlan; /* VLAN if stripped */
285 } __packed rxd;
286
287 __le32 vals[2];
288 };
289 };
290
291 #define NFP_NET_META_FIELD_MASK GENMASK(NFP_NET_META_FIELD_SIZE - 1, 0)
292
293 struct nfp_meta_parsed {
294 u8 hash_type;
295 u8 csum_type;
296 u32 hash;
297 u32 mark;
298 u32 portid;
299 __wsum csum;
300 };
301
302 struct nfp_net_rx_hash {
303 __be32 hash_type;
304 __be32 hash;
305 };
306
307 /**
308 * struct nfp_net_rx_buf - software RX buffer descriptor
309 * @frag: page fragment buffer
310 * @dma_addr: DMA mapping address of the buffer
311 */
312 struct nfp_net_rx_buf {
313 void *frag;
314 dma_addr_t dma_addr;
315 };
316
317 /**
318 * struct nfp_net_rx_ring - RX ring structure
319 * @r_vec: Back pointer to ring vector structure
320 * @cnt: Size of the queue in number of descriptors
321 * @wr_p: FL/RX ring write pointer (free running)
322 * @rd_p: FL/RX ring read pointer (free running)
323 * @idx: Ring index from Linux's perspective
324 * @fl_qcidx: Queue Controller Peripheral (QCP) queue index for the freelist
325 * @qcp_fl: Pointer to base of the QCP freelist queue
326 * @rxbufs: Array of transmitted FL/RX buffers
327 * @rxds: Virtual address of FL/RX ring in host memory
328 * @xdp_rxq: RX-ring info avail for XDP
329 * @dma: DMA address of the FL/RX ring
330 * @size: Size, in bytes, of the FL/RX ring (needed to free)
331 */
332 struct nfp_net_rx_ring {
333 struct nfp_net_r_vector *r_vec;
334
335 u32 cnt;
336 u32 wr_p;
337 u32 rd_p;
338
339 u32 idx;
340
341 int fl_qcidx;
342 u8 __iomem *qcp_fl;
343
344 struct nfp_net_rx_buf *rxbufs;
345 struct nfp_net_rx_desc *rxds;
346
347 struct xdp_rxq_info xdp_rxq;
348
349 dma_addr_t dma;
350 size_t size;
351 } ____cacheline_aligned;
352
353 /**
354 * struct nfp_net_r_vector - Per ring interrupt vector configuration
355 * @nfp_net: Backpointer to nfp_net structure
356 * @napi: NAPI structure for this ring vec
357 * @tasklet: ctrl vNIC, tasklet for servicing the r_vec
358 * @queue: ctrl vNIC, send queue
359 * @lock: ctrl vNIC, r_vec lock protects @queue
360 * @tx_ring: Pointer to TX ring
361 * @rx_ring: Pointer to RX ring
362 * @xdp_ring: Pointer to an extra TX ring for XDP
363 * @irq_entry: MSI-X table entry (use for talking to the device)
364 * @event_ctr: Number of interrupt
365 * @rx_dim: Dynamic interrupt moderation structure for RX
366 * @tx_dim: Dynamic interrupt moderation structure for TX
367 * @rx_sync: Seqlock for atomic updates of RX stats
368 * @rx_pkts: Number of received packets
369 * @rx_bytes: Number of received bytes
370 * @rx_drops: Number of packets dropped on RX due to lack of resources
371 * @hw_csum_rx_ok: Counter of packets where the HW checksum was OK
372 * @hw_csum_rx_inner_ok: Counter of packets where the inner HW checksum was OK
373 * @hw_csum_rx_complete: Counter of packets with CHECKSUM_COMPLETE reported
374 * @hw_csum_rx_error: Counter of packets with bad checksums
375 * @hw_tls_rx: Number of packets with TLS decrypted by hardware
376 * @tx_sync: Seqlock for atomic updates of TX stats
377 * @tx_pkts: Number of Transmitted packets
378 * @tx_bytes: Number of Transmitted bytes
379 * @hw_csum_tx: Counter of packets with TX checksum offload requested
380 * @hw_csum_tx_inner: Counter of inner TX checksum offload requests
381 * @tx_gather: Counter of packets with Gather DMA
382 * @tx_lso: Counter of LSO packets sent
383 * @hw_tls_tx: Counter of TLS packets sent with crypto offloaded to HW
384 * @tls_tx_fallback: Counter of TLS packets sent which had to be encrypted
385 * by the fallback path because packets came out of order
386 * @tls_tx_no_fallback: Counter of TLS packets not sent because the fallback
387 * path could not encrypt them
388 * @tx_errors: How many TX errors were encountered
389 * @tx_busy: How often was TX busy (no space)?
390 * @rx_replace_buf_alloc_fail: Counter of RX buffer allocation failures
391 * @irq_vector: Interrupt vector number (use for talking to the OS)
392 * @handler: Interrupt handler for this ring vector
393 * @name: Name of the interrupt vector
394 * @affinity_mask: SMP affinity mask for this vector
395 *
396 * This structure ties RX and TX rings to interrupt vectors and a NAPI
397 * context. This currently only supports one RX and TX ring per
398 * interrupt vector but might be extended in the future to allow
399 * association of multiple rings per vector.
400 */
401 struct nfp_net_r_vector {
402 struct nfp_net *nfp_net;
403 union {
404 struct napi_struct napi;
405 struct {
406 struct tasklet_struct tasklet;
407 struct sk_buff_head queue;
408 spinlock_t lock;
409 };
410 };
411
412 struct nfp_net_tx_ring *tx_ring;
413 struct nfp_net_rx_ring *rx_ring;
414
415 u16 irq_entry;
416
417 u16 event_ctr;
418 struct dim rx_dim;
419 struct dim tx_dim;
420
421 struct u64_stats_sync rx_sync;
422 u64 rx_pkts;
423 u64 rx_bytes;
424 u64 rx_drops;
425 u64 hw_csum_rx_ok;
426 u64 hw_csum_rx_inner_ok;
427 u64 hw_csum_rx_complete;
428 u64 hw_tls_rx;
429
430 u64 hw_csum_rx_error;
431 u64 rx_replace_buf_alloc_fail;
432
433 struct nfp_net_tx_ring *xdp_ring;
434
435 struct u64_stats_sync tx_sync;
436 u64 tx_pkts;
437 u64 tx_bytes;
438
439 u64 ____cacheline_aligned_in_smp hw_csum_tx;
440 u64 hw_csum_tx_inner;
441 u64 tx_gather;
442 u64 tx_lso;
443 u64 hw_tls_tx;
444
445 u64 tls_tx_fallback;
446 u64 tls_tx_no_fallback;
447 u64 tx_errors;
448 u64 tx_busy;
449
450 /* Cold data follows */
451
452 u32 irq_vector;
453 irq_handler_t handler;
454 char name[IFNAMSIZ + 8];
455 cpumask_t affinity_mask;
456 } ____cacheline_aligned;
457
458 /* Firmware version as it is written in the 32bit value in the BAR */
459 struct nfp_net_fw_version {
460 u8 minor;
461 u8 major;
462 u8 class;
463 u8 resv;
464 } __packed;
465
nfp_net_fw_ver_eq(struct nfp_net_fw_version * fw_ver,u8 resv,u8 class,u8 major,u8 minor)466 static inline bool nfp_net_fw_ver_eq(struct nfp_net_fw_version *fw_ver,
467 u8 resv, u8 class, u8 major, u8 minor)
468 {
469 return fw_ver->resv == resv &&
470 fw_ver->class == class &&
471 fw_ver->major == major &&
472 fw_ver->minor == minor;
473 }
474
475 struct nfp_stat_pair {
476 u64 pkts;
477 u64 bytes;
478 };
479
480 /**
481 * struct nfp_net_dp - NFP network device datapath data structure
482 * @dev: Backpointer to struct device
483 * @netdev: Backpointer to net_device structure
484 * @is_vf: Is the driver attached to a VF?
485 * @chained_metadata_format: Firemware will use new metadata format
486 * @ktls_tx: Is kTLS TX enabled?
487 * @rx_dma_dir: Mapping direction for RX buffers
488 * @rx_dma_off: Offset at which DMA packets (for XDP headroom)
489 * @rx_offset: Offset in the RX buffers where packet data starts
490 * @ctrl: Local copy of the control register/word.
491 * @fl_bufsz: Currently configured size of the freelist buffers
492 * @xdp_prog: Installed XDP program
493 * @tx_rings: Array of pre-allocated TX ring structures
494 * @rx_rings: Array of pre-allocated RX ring structures
495 * @ctrl_bar: Pointer to mapped control BAR
496 *
497 * @txd_cnt: Size of the TX ring in number of descriptors
498 * @rxd_cnt: Size of the RX ring in number of descriptors
499 * @num_r_vecs: Number of used ring vectors
500 * @num_tx_rings: Currently configured number of TX rings
501 * @num_stack_tx_rings: Number of TX rings used by the stack (not XDP)
502 * @num_rx_rings: Currently configured number of RX rings
503 * @mtu: Device MTU
504 */
505 struct nfp_net_dp {
506 struct device *dev;
507 struct net_device *netdev;
508
509 u8 is_vf:1;
510 u8 chained_metadata_format:1;
511 u8 ktls_tx:1;
512
513 u8 rx_dma_dir;
514 u8 rx_offset;
515
516 u32 rx_dma_off;
517
518 u32 ctrl;
519 u32 fl_bufsz;
520
521 struct bpf_prog *xdp_prog;
522
523 struct nfp_net_tx_ring *tx_rings;
524 struct nfp_net_rx_ring *rx_rings;
525
526 u8 __iomem *ctrl_bar;
527
528 /* Cold data follows */
529
530 unsigned int txd_cnt;
531 unsigned int rxd_cnt;
532
533 unsigned int num_r_vecs;
534
535 unsigned int num_tx_rings;
536 unsigned int num_stack_tx_rings;
537 unsigned int num_rx_rings;
538
539 unsigned int mtu;
540 };
541
542 /**
543 * struct nfp_net - NFP network device structure
544 * @dp: Datapath structure
545 * @id: vNIC id within the PF (0 for VFs)
546 * @fw_ver: Firmware version
547 * @cap: Capabilities advertised by the Firmware
548 * @max_mtu: Maximum support MTU advertised by the Firmware
549 * @rss_hfunc: RSS selected hash function
550 * @rss_cfg: RSS configuration
551 * @rss_key: RSS secret key
552 * @rss_itbl: RSS indirection table
553 * @xdp: Information about the driver XDP program
554 * @xdp_hw: Information about the HW XDP program
555 * @max_r_vecs: Number of allocated interrupt vectors for RX/TX
556 * @max_tx_rings: Maximum number of TX rings supported by the Firmware
557 * @max_rx_rings: Maximum number of RX rings supported by the Firmware
558 * @stride_rx: Queue controller RX queue spacing
559 * @stride_tx: Queue controller TX queue spacing
560 * @r_vecs: Pre-allocated array of ring vectors
561 * @irq_entries: Pre-allocated array of MSI-X entries
562 * @lsc_handler: Handler for Link State Change interrupt
563 * @lsc_name: Name for Link State Change interrupt
564 * @exn_handler: Handler for Exception interrupt
565 * @exn_name: Name for Exception interrupt
566 * @shared_handler: Handler for shared interrupts
567 * @shared_name: Name for shared interrupt
568 * @me_freq_mhz: ME clock_freq (MHz)
569 * @reconfig_lock: Protects @reconfig_posted, @reconfig_timer_active,
570 * @reconfig_sync_present and HW reconfiguration request
571 * regs/machinery from async requests (sync must take
572 * @bar_lock)
573 * @reconfig_posted: Pending reconfig bits coming from async sources
574 * @reconfig_timer_active: Timer for reading reconfiguration results is pending
575 * @reconfig_sync_present: Some thread is performing synchronous reconfig
576 * @reconfig_timer: Timer for async reading of reconfig results
577 * @reconfig_in_progress_update: Update FW is processing now (debug only)
578 * @bar_lock: vNIC config BAR access lock, protects: update,
579 * mailbox area, crypto TLV
580 * @link_up: Is the link up?
581 * @link_status_lock: Protects @link_* and ensures atomicity with BAR reading
582 * @rx_coalesce_adapt_on: Is RX interrupt moderation adaptive?
583 * @tx_coalesce_adapt_on: Is TX interrupt moderation adaptive?
584 * @rx_coalesce_usecs: RX interrupt moderation usecs delay parameter
585 * @rx_coalesce_max_frames: RX interrupt moderation frame count parameter
586 * @tx_coalesce_usecs: TX interrupt moderation usecs delay parameter
587 * @tx_coalesce_max_frames: TX interrupt moderation frame count parameter
588 * @qcp_cfg: Pointer to QCP queue used for configuration notification
589 * @tx_bar: Pointer to mapped TX queues
590 * @rx_bar: Pointer to mapped FL/RX queues
591 * @tlv_caps: Parsed TLV capabilities
592 * @ktls_tx_conn_cnt: Number of offloaded kTLS TX connections
593 * @ktls_rx_conn_cnt: Number of offloaded kTLS RX connections
594 * @ktls_conn_id_gen: Trivial generator for kTLS connection ids (for TX)
595 * @ktls_no_space: Counter of firmware rejecting kTLS connection due to
596 * lack of space
597 * @ktls_rx_resync_req: Counter of TLS RX resync requested
598 * @ktls_rx_resync_ign: Counter of TLS RX resync requests ignored
599 * @ktls_rx_resync_sent: Counter of TLS RX resync completed
600 * @mbox_cmsg: Common Control Message via vNIC mailbox state
601 * @mbox_cmsg.queue: CCM mbox queue of pending messages
602 * @mbox_cmsg.wq: CCM mbox wait queue of waiting processes
603 * @mbox_cmsg.workq: CCM mbox work queue for @wait_work and @runq_work
604 * @mbox_cmsg.wait_work: CCM mbox posted msg reconfig wait work
605 * @mbox_cmsg.runq_work: CCM mbox posted msg queue runner work
606 * @mbox_cmsg.tag: CCM mbox message tag allocator
607 * @debugfs_dir: Device directory in debugfs
608 * @vnic_list: Entry on device vNIC list
609 * @pdev: Backpointer to PCI device
610 * @app: APP handle if available
611 * @vnic_no_name: For non-port PF vNIC make ndo_get_phys_port_name return
612 * -EOPNOTSUPP to keep backwards compatibility (set by app)
613 * @port: Pointer to nfp_port structure if vNIC is a port
614 * @app_priv: APP private data for this vNIC
615 */
616 struct nfp_net {
617 struct nfp_net_dp dp;
618
619 struct nfp_net_fw_version fw_ver;
620
621 u32 id;
622
623 u32 cap;
624 u32 max_mtu;
625
626 u8 rss_hfunc;
627 u32 rss_cfg;
628 u8 rss_key[NFP_NET_CFG_RSS_KEY_SZ];
629 u8 rss_itbl[NFP_NET_CFG_RSS_ITBL_SZ];
630
631 struct xdp_attachment_info xdp;
632 struct xdp_attachment_info xdp_hw;
633
634 unsigned int max_tx_rings;
635 unsigned int max_rx_rings;
636
637 int stride_tx;
638 int stride_rx;
639
640 unsigned int max_r_vecs;
641 struct nfp_net_r_vector r_vecs[NFP_NET_MAX_R_VECS];
642 struct msix_entry irq_entries[NFP_NET_MAX_IRQS];
643
644 irq_handler_t lsc_handler;
645 char lsc_name[IFNAMSIZ + 8];
646
647 irq_handler_t exn_handler;
648 char exn_name[IFNAMSIZ + 8];
649
650 irq_handler_t shared_handler;
651 char shared_name[IFNAMSIZ + 8];
652
653 u32 me_freq_mhz;
654
655 bool link_up;
656 spinlock_t link_status_lock;
657
658 spinlock_t reconfig_lock;
659 u32 reconfig_posted;
660 bool reconfig_timer_active;
661 bool reconfig_sync_present;
662 struct timer_list reconfig_timer;
663 u32 reconfig_in_progress_update;
664
665 struct semaphore bar_lock;
666
667 bool rx_coalesce_adapt_on;
668 bool tx_coalesce_adapt_on;
669 u32 rx_coalesce_usecs;
670 u32 rx_coalesce_max_frames;
671 u32 tx_coalesce_usecs;
672 u32 tx_coalesce_max_frames;
673
674 u8 __iomem *qcp_cfg;
675
676 u8 __iomem *tx_bar;
677 u8 __iomem *rx_bar;
678
679 struct nfp_net_tlv_caps tlv_caps;
680
681 unsigned int ktls_tx_conn_cnt;
682 unsigned int ktls_rx_conn_cnt;
683
684 atomic64_t ktls_conn_id_gen;
685
686 atomic_t ktls_no_space;
687 atomic_t ktls_rx_resync_req;
688 atomic_t ktls_rx_resync_ign;
689 atomic_t ktls_rx_resync_sent;
690
691 struct {
692 struct sk_buff_head queue;
693 wait_queue_head_t wq;
694 struct workqueue_struct *workq;
695 struct work_struct wait_work;
696 struct work_struct runq_work;
697 u16 tag;
698 } mbox_cmsg;
699
700 struct dentry *debugfs_dir;
701
702 struct list_head vnic_list;
703
704 struct pci_dev *pdev;
705 struct nfp_app *app;
706
707 bool vnic_no_name;
708
709 struct nfp_port *port;
710
711 void *app_priv;
712 };
713
714 /* Functions to read/write from/to a BAR
715 * Performs any endian conversion necessary.
716 */
nn_readb(struct nfp_net * nn,int off)717 static inline u16 nn_readb(struct nfp_net *nn, int off)
718 {
719 return readb(nn->dp.ctrl_bar + off);
720 }
721
nn_writeb(struct nfp_net * nn,int off,u8 val)722 static inline void nn_writeb(struct nfp_net *nn, int off, u8 val)
723 {
724 writeb(val, nn->dp.ctrl_bar + off);
725 }
726
nn_readw(struct nfp_net * nn,int off)727 static inline u16 nn_readw(struct nfp_net *nn, int off)
728 {
729 return readw(nn->dp.ctrl_bar + off);
730 }
731
nn_writew(struct nfp_net * nn,int off,u16 val)732 static inline void nn_writew(struct nfp_net *nn, int off, u16 val)
733 {
734 writew(val, nn->dp.ctrl_bar + off);
735 }
736
nn_readl(struct nfp_net * nn,int off)737 static inline u32 nn_readl(struct nfp_net *nn, int off)
738 {
739 return readl(nn->dp.ctrl_bar + off);
740 }
741
nn_writel(struct nfp_net * nn,int off,u32 val)742 static inline void nn_writel(struct nfp_net *nn, int off, u32 val)
743 {
744 writel(val, nn->dp.ctrl_bar + off);
745 }
746
nn_readq(struct nfp_net * nn,int off)747 static inline u64 nn_readq(struct nfp_net *nn, int off)
748 {
749 return readq(nn->dp.ctrl_bar + off);
750 }
751
nn_writeq(struct nfp_net * nn,int off,u64 val)752 static inline void nn_writeq(struct nfp_net *nn, int off, u64 val)
753 {
754 writeq(val, nn->dp.ctrl_bar + off);
755 }
756
757 /* Flush posted PCI writes by reading something without side effects */
nn_pci_flush(struct nfp_net * nn)758 static inline void nn_pci_flush(struct nfp_net *nn)
759 {
760 nn_readl(nn, NFP_NET_CFG_VERSION);
761 }
762
763 /* Queue Controller Peripheral access functions and definitions.
764 *
765 * Some of the BARs of the NFP are mapped to portions of the Queue
766 * Controller Peripheral (QCP) address space on the NFP. A QCP queue
767 * has a read and a write pointer (as well as a size and flags,
768 * indicating overflow etc). The QCP offers a number of different
769 * operation on queue pointers, but here we only offer function to
770 * either add to a pointer or to read the pointer value.
771 */
772 #define NFP_QCP_QUEUE_ADDR_SZ 0x800
773 #define NFP_QCP_QUEUE_AREA_SZ 0x80000
774 #define NFP_QCP_QUEUE_OFF(_x) ((_x) * NFP_QCP_QUEUE_ADDR_SZ)
775 #define NFP_QCP_QUEUE_ADD_RPTR 0x0000
776 #define NFP_QCP_QUEUE_ADD_WPTR 0x0004
777 #define NFP_QCP_QUEUE_STS_LO 0x0008
778 #define NFP_QCP_QUEUE_STS_LO_READPTR_mask 0x3ffff
779 #define NFP_QCP_QUEUE_STS_HI 0x000c
780 #define NFP_QCP_QUEUE_STS_HI_WRITEPTR_mask 0x3ffff
781
782 /* The offset of a QCP queues in the PCIe Target */
783 #define NFP_PCIE_QUEUE(_q) (0x80000 + (NFP_QCP_QUEUE_ADDR_SZ * ((_q) & 0xff)))
784
785 /* nfp_qcp_ptr - Read or Write Pointer of a queue */
786 enum nfp_qcp_ptr {
787 NFP_QCP_READ_PTR = 0,
788 NFP_QCP_WRITE_PTR
789 };
790
791 /* There appear to be an *undocumented* upper limit on the value which
792 * one can add to a queue and that value is either 0x3f or 0x7f. We
793 * go with 0x3f as a conservative measure.
794 */
795 #define NFP_QCP_MAX_ADD 0x3f
796
_nfp_qcp_ptr_add(u8 __iomem * q,enum nfp_qcp_ptr ptr,u32 val)797 static inline void _nfp_qcp_ptr_add(u8 __iomem *q,
798 enum nfp_qcp_ptr ptr, u32 val)
799 {
800 u32 off;
801
802 if (ptr == NFP_QCP_READ_PTR)
803 off = NFP_QCP_QUEUE_ADD_RPTR;
804 else
805 off = NFP_QCP_QUEUE_ADD_WPTR;
806
807 while (val > NFP_QCP_MAX_ADD) {
808 writel(NFP_QCP_MAX_ADD, q + off);
809 val -= NFP_QCP_MAX_ADD;
810 }
811
812 writel(val, q + off);
813 }
814
815 /**
816 * nfp_qcp_rd_ptr_add() - Add the value to the read pointer of a queue
817 *
818 * @q: Base address for queue structure
819 * @val: Value to add to the queue pointer
820 *
821 * If @val is greater than @NFP_QCP_MAX_ADD multiple writes are performed.
822 */
nfp_qcp_rd_ptr_add(u8 __iomem * q,u32 val)823 static inline void nfp_qcp_rd_ptr_add(u8 __iomem *q, u32 val)
824 {
825 _nfp_qcp_ptr_add(q, NFP_QCP_READ_PTR, val);
826 }
827
828 /**
829 * nfp_qcp_wr_ptr_add() - Add the value to the write pointer of a queue
830 *
831 * @q: Base address for queue structure
832 * @val: Value to add to the queue pointer
833 *
834 * If @val is greater than @NFP_QCP_MAX_ADD multiple writes are performed.
835 */
nfp_qcp_wr_ptr_add(u8 __iomem * q,u32 val)836 static inline void nfp_qcp_wr_ptr_add(u8 __iomem *q, u32 val)
837 {
838 _nfp_qcp_ptr_add(q, NFP_QCP_WRITE_PTR, val);
839 }
840
_nfp_qcp_read(u8 __iomem * q,enum nfp_qcp_ptr ptr)841 static inline u32 _nfp_qcp_read(u8 __iomem *q, enum nfp_qcp_ptr ptr)
842 {
843 u32 off;
844 u32 val;
845
846 if (ptr == NFP_QCP_READ_PTR)
847 off = NFP_QCP_QUEUE_STS_LO;
848 else
849 off = NFP_QCP_QUEUE_STS_HI;
850
851 val = readl(q + off);
852
853 if (ptr == NFP_QCP_READ_PTR)
854 return val & NFP_QCP_QUEUE_STS_LO_READPTR_mask;
855 else
856 return val & NFP_QCP_QUEUE_STS_HI_WRITEPTR_mask;
857 }
858
859 /**
860 * nfp_qcp_rd_ptr_read() - Read the current read pointer value for a queue
861 * @q: Base address for queue structure
862 *
863 * Return: Value read.
864 */
nfp_qcp_rd_ptr_read(u8 __iomem * q)865 static inline u32 nfp_qcp_rd_ptr_read(u8 __iomem *q)
866 {
867 return _nfp_qcp_read(q, NFP_QCP_READ_PTR);
868 }
869
870 /**
871 * nfp_qcp_wr_ptr_read() - Read the current write pointer value for a queue
872 * @q: Base address for queue structure
873 *
874 * Return: Value read.
875 */
nfp_qcp_wr_ptr_read(u8 __iomem * q)876 static inline u32 nfp_qcp_wr_ptr_read(u8 __iomem *q)
877 {
878 return _nfp_qcp_read(q, NFP_QCP_WRITE_PTR);
879 }
880
nfp_net_is_data_vnic(struct nfp_net * nn)881 static inline bool nfp_net_is_data_vnic(struct nfp_net *nn)
882 {
883 WARN_ON_ONCE(!nn->dp.netdev && nn->port);
884 return !!nn->dp.netdev;
885 }
886
nfp_net_running(struct nfp_net * nn)887 static inline bool nfp_net_running(struct nfp_net *nn)
888 {
889 return nn->dp.ctrl & NFP_NET_CFG_CTRL_ENABLE;
890 }
891
nfp_net_name(struct nfp_net * nn)892 static inline const char *nfp_net_name(struct nfp_net *nn)
893 {
894 return nn->dp.netdev ? nn->dp.netdev->name : "ctrl";
895 }
896
nfp_ctrl_lock(struct nfp_net * nn)897 static inline void nfp_ctrl_lock(struct nfp_net *nn)
898 __acquires(&nn->r_vecs[0].lock)
899 {
900 spin_lock_bh(&nn->r_vecs[0].lock);
901 }
902
nfp_ctrl_unlock(struct nfp_net * nn)903 static inline void nfp_ctrl_unlock(struct nfp_net *nn)
904 __releases(&nn->r_vecs[0].lock)
905 {
906 spin_unlock_bh(&nn->r_vecs[0].lock);
907 }
908
nn_ctrl_bar_lock(struct nfp_net * nn)909 static inline void nn_ctrl_bar_lock(struct nfp_net *nn)
910 {
911 down(&nn->bar_lock);
912 }
913
nn_ctrl_bar_trylock(struct nfp_net * nn)914 static inline bool nn_ctrl_bar_trylock(struct nfp_net *nn)
915 {
916 return !down_trylock(&nn->bar_lock);
917 }
918
nn_ctrl_bar_unlock(struct nfp_net * nn)919 static inline void nn_ctrl_bar_unlock(struct nfp_net *nn)
920 {
921 up(&nn->bar_lock);
922 }
923
924 /* Globals */
925 extern const char nfp_driver_version[];
926
927 extern const struct net_device_ops nfp_net_netdev_ops;
928
nfp_netdev_is_nfp_net(struct net_device * netdev)929 static inline bool nfp_netdev_is_nfp_net(struct net_device *netdev)
930 {
931 return netdev->netdev_ops == &nfp_net_netdev_ops;
932 }
933
nfp_net_coalesce_para_check(u32 usecs,u32 pkts)934 static inline int nfp_net_coalesce_para_check(u32 usecs, u32 pkts)
935 {
936 if ((usecs >= ((1 << 16) - 1)) || (pkts >= ((1 << 16) - 1)))
937 return -EINVAL;
938
939 return 0;
940 }
941
942 /* Prototypes */
943 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver,
944 void __iomem *ctrl_bar);
945
946 struct nfp_net *
947 nfp_net_alloc(struct pci_dev *pdev, void __iomem *ctrl_bar, bool needs_netdev,
948 unsigned int max_tx_rings, unsigned int max_rx_rings);
949 void nfp_net_free(struct nfp_net *nn);
950
951 int nfp_net_init(struct nfp_net *nn);
952 void nfp_net_clean(struct nfp_net *nn);
953
954 int nfp_ctrl_open(struct nfp_net *nn);
955 void nfp_ctrl_close(struct nfp_net *nn);
956
957 void nfp_net_set_ethtool_ops(struct net_device *netdev);
958 void nfp_net_info(struct nfp_net *nn);
959 int __nfp_net_reconfig(struct nfp_net *nn, u32 update);
960 int nfp_net_reconfig(struct nfp_net *nn, u32 update);
961 unsigned int nfp_net_rss_key_sz(struct nfp_net *nn);
962 void nfp_net_rss_write_itbl(struct nfp_net *nn);
963 void nfp_net_rss_write_key(struct nfp_net *nn);
964 void nfp_net_coalesce_write_cfg(struct nfp_net *nn);
965 int nfp_net_mbox_lock(struct nfp_net *nn, unsigned int data_size);
966 int nfp_net_mbox_reconfig(struct nfp_net *nn, u32 mbox_cmd);
967 int nfp_net_mbox_reconfig_and_unlock(struct nfp_net *nn, u32 mbox_cmd);
968 void nfp_net_mbox_reconfig_post(struct nfp_net *nn, u32 update);
969 int nfp_net_mbox_reconfig_wait_posted(struct nfp_net *nn);
970
971 unsigned int
972 nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries,
973 unsigned int min_irqs, unsigned int want_irqs);
974 void nfp_net_irqs_disable(struct pci_dev *pdev);
975 void
976 nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries,
977 unsigned int n);
978
979 struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn);
980 int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *new,
981 struct netlink_ext_ack *extack);
982
983 #ifdef CONFIG_NFP_DEBUG
984 void nfp_net_debugfs_create(void);
985 void nfp_net_debugfs_destroy(void);
986 struct dentry *nfp_net_debugfs_device_add(struct pci_dev *pdev);
987 void nfp_net_debugfs_vnic_add(struct nfp_net *nn, struct dentry *ddir);
988 void nfp_net_debugfs_dir_clean(struct dentry **dir);
989 #else
nfp_net_debugfs_create(void)990 static inline void nfp_net_debugfs_create(void)
991 {
992 }
993
nfp_net_debugfs_destroy(void)994 static inline void nfp_net_debugfs_destroy(void)
995 {
996 }
997
nfp_net_debugfs_device_add(struct pci_dev * pdev)998 static inline struct dentry *nfp_net_debugfs_device_add(struct pci_dev *pdev)
999 {
1000 return NULL;
1001 }
1002
1003 static inline void
nfp_net_debugfs_vnic_add(struct nfp_net * nn,struct dentry * ddir)1004 nfp_net_debugfs_vnic_add(struct nfp_net *nn, struct dentry *ddir)
1005 {
1006 }
1007
nfp_net_debugfs_dir_clean(struct dentry ** dir)1008 static inline void nfp_net_debugfs_dir_clean(struct dentry **dir)
1009 {
1010 }
1011 #endif /* CONFIG_NFP_DEBUG */
1012
1013 #endif /* _NFP_NET_H_ */
1014
